Time evolution of dynamic systems is governed by the interactions between system variables. The interactions can be directional, (non)linear, and noisy, and the evolution scheme can be either deterministic or stochastic. Here, we usually meet an inverse problem to infer the interactions given time series data of system variables. A well-known example is the reconstruction of neural circuits from brain signals. In this seminar, I will introduce a new method based on statistical physics, which is very powerful to solve the network reconstruction even with little data.